Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dive into Time-Series Anomaly Detection: A Decade Review

Paul Boniol, Qinghua Liu, Mingyi Huang, Themis Palpanas, John Paparrizos

TL;DR

This survey tackles time-series anomaly detection by proposing a process-centric taxonomy that unifies distance-, density-, and prediction-based approaches. It systematically reviews methods from KNN and LOF to matrix profile, graph- and grammar-based representations, and forecasting- and reconstruction-based models, while summarizing benchmarks and evaluation practices. A meta-analysis shows a surge in deep-learning–driven methods after 2016, a tilt toward univariate subsequence detection, and persistent gaps in streaming, missing-value, and multivariate scenarios. The work highlights the need for standardized benchmarks and threshold-independent evaluation to fairly compare methods across diverse real-world datasets, guiding future research toward robust, scalable, and auto-configurable anomaly detectors.

Abstract

Recent advances in data collection technology, accompanied by the ever-rising volume and velocity of streaming data, underscore the vital need for time series analytics. In this regard, time-series anomaly detection has been an important activity, entailing various applications in fields such as cyber security, financial markets, law enforcement, and health care. While traditional literature on anomaly detection is centered on statistical measures, the increasing number of machine learning algorithms in recent years call for a structured, general characterization of the research methods for time-series anomaly detection. This survey groups and summarizes anomaly detection existing solutions under a process-centric taxonomy in the time series context. In addition to giving an original categorization of anomaly detection methods, we also perform a meta-analysis of the literature and outline general trends in time-series anomaly detection research.

Dive into Time-Series Anomaly Detection: A Decade Review

TL;DR

This survey tackles time-series anomaly detection by proposing a process-centric taxonomy that unifies distance-, density-, and prediction-based approaches. It systematically reviews methods from KNN and LOF to matrix profile, graph- and grammar-based representations, and forecasting- and reconstruction-based models, while summarizing benchmarks and evaluation practices. A meta-analysis shows a surge in deep-learning–driven methods after 2016, a tilt toward univariate subsequence detection, and persistent gaps in streaming, missing-value, and multivariate scenarios. The work highlights the need for standardized benchmarks and threshold-independent evaluation to fairly compare methods across diverse real-world datasets, guiding future research toward robust, scalable, and auto-configurable anomaly detectors.

Abstract

Recent advances in data collection technology, accompanied by the ever-rising volume and velocity of streaming data, underscore the vital need for time series analytics. In this regard, time-series anomaly detection has been an important activity, entailing various applications in fields such as cyber security, financial markets, law enforcement, and health care. While traditional literature on anomaly detection is centered on statistical measures, the increasing number of machine learning algorithms in recent years call for a structured, general characterization of the research methods for time-series anomaly detection. This survey groups and summarizes anomaly detection existing solutions under a process-centric taxonomy in the time series context. In addition to giving an original categorization of anomaly detection methods, we also perform a meta-analysis of the literature and outline general trends in time-series anomaly detection research.
Paper Structure (60 sections, 19 equations, 22 figures, 4 tables)

This paper contains 60 sections, 19 equations, 22 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Time-Series Anomaly Detection Overview
  3. On the Definition of Anomalies in Time Series
  4. Types of Anomalies in Time Series
  5. Univariate versus Multivariate
  6. Unsupervised, Semi-supervised versus Supervised
  7. Anomaly Detection Pipelines
  8. Anomaly Detection Taxonomy
  9. Distance-based
  10. Density-based
  11. Prediction-based
  12. Scoring Process
  13. Survey Organization
  14. Time Series Notations
  15. Distance-based Methods
  16. ...and 45 more sections

Figures (22)

  • Figure 1: Examples of different time series applications and types of anomalies.
  • Figure 2: Synthetic illustration of the three time series anomaly types: (a) point; (b) contextual; and (c) collective anomalies.
  • Figure 3: Synthetic example comparing anomalies in univariate and multivariate time series for (a) a point outlier and (b) a sequence outlier.
  • Figure 4: Time series anomaly detection pipeline.
  • Figure 5: Process-centric anomaly detection taxonomy.
  • ...and 17 more figures

Theorems & Definitions (5)

  • definition 1: Top-k $m^{th}$-discord
  • definition 2: Matrix Profile
  • definition 3: Matrix Profile Index
  • definition 4: Time Series Self-Join
  • definition 5: Time Series Join